Forum - View topicNEWS: DLsite, Ci-en, pixiv FANBOX, Fantia Ban AI-Generated Content
Goto page Previous 1, 2 Note: this is the discussion thread for this article |
Author | Message | |||
---|---|---|---|---|
tsog
Posts: 228 |
|
|||
As the sites themselves probably understand, banning generative AI works is only a stopgap (temporary) solution at best.
Given the popularity of AI anime art on Twitter/Pixiv (and more importantly, all for the price of FREE for the viewers), there are plenty of eyeballs for AI art that won't be going away any time soon, if ever. Paying for AI art may sound wrong at first glance, but it may simply be that the payers don't have the compute hardware to generate said art, or that they just enjoy AI art that someone else generated with their own secret prompts. It's somewhat like paying to watch someone else play a game instead of playing the game yourself, an idea that I admittedly found baffling at first. Democratization of AI computing could discourage a lot of the payers. If Stable Diffusion can run on smartphones that most people have, there would be less demand for someone else's generations. My biggest issue with generative AI is the lack of clarity around copyright. Laws need to be updated to clarify whether training data falls under fair use or not, and what recourses human artists have when their art have been heavily "borrowed" by AI works. |
||||
FeelMyBlade
Posts: 142 |
|
|||
My only qualm with AI art is it's all got that semi-realistic filter on it. Once they can get rid of that and make it look like traditional 2D art, then that's going to be when things get really interesting. |
||||
Kicksville
Posts: 1180 |
|
|||
The problem is that many (most?) casual users don't know about that setting. We're talking about directly monetized works on FANBOX, so going straight to banning AI seems to be how they've decided to deal with sensitivities there. But for Pixiv in general, it seems they're still trying to accommodate everyone - in their post about how they'll be dealing with AI on Pixiv, they mention updating the AI filter setting to be easier to access. https://www.pixiv.net/info.php?id=9524 Also mentions looking into dealing with mimics of specific user styles, single AI uploaders slamming out high volumes of content, etc. |
||||
tfwnoymir
Posts: 315 Location: Hungary |
|
|||
You're a lifesaver! I had no idea they've made this change in the settings up until now, not that I check that out all the time. |
||||
Richmyster84
Posts: 174 |
|
|||
I don't mind a lot of the AI art I've seen so far. The big problem I have is how repetitive the art is. The major issue I have is that some AI artists upload HUNDREDS of images a day which has to be a strain on these websites' servers.
These images aren't being bought so not only does it not make any money for the websites they're also bogging down server space which costs the websites money while also slowing down the download/upload speed for all users. Pixiv has been running noticeably slower for the past few days imo. I think a good compromise would be that individual artists can only upload a certain number of AI generated images a day or over a period of time equal to how often it would take an actual artist to create the same amount of artwork. That would solve the over influx of subpar material being uploaded (essentially forcing AI artists to only upload their best work) while also allowing AI artists a space on these websites. |
||||
FeelMyBlade
Posts: 142 |
|
|||
That would be like if Twitch or YouTube said no one with less than a certain amount of viewers can stream on their website or upload videos. It's technically true when it comes to bandwidth cost, but it's one of those things where once you start down the slippery slope it seems unfair to try to justify where and why you drew the line at. There's a lot of bad art on pixiv that's worse than the average AI picture so I'm not sure a quality control method would be fair. |
||||
ViviP
Posts: 73 |
|
|||
FANBOX Announcement (now in english) https://www.fanbox.cc/@official-en/posts/5934381
Pixiv Announcment https://www.pixiv.net/info.php?id=9524 I think Fanbox themselves explain their reasoning perfectly:
Seems sensible me. As for the Pixiv Announcement:
Seems a bit flimsy to me. It's nice that Pixiv is acknowledging user concerns, but It seems like they really don't have any effective ways to deal with those concerns. |
||||
AQuin1904
Posts: 266 |
|
|||
Good. The sheer firehose of "AI" content from people hoping to make a quick buck was a serious blow to usability. We've seen how bad this can get with asset-flips and other low-effort titles on video game marketplaces, and I really don't need art sites falling off the same cliff.
Anyone making ML artwork good enough to not be detectable is probably also spending enough time on refinements and manual revisions that they aren't flooding the marketplace, so they won't be as much of a problem. |
||||
AsleepBySunset
Posts: 208 |
|
|||
The idea there's going to be a mass exodus from art websites which ban AI images is ridiculous, if all AI image creators left a social media platform I created, I'd be delighted, and the idea that regular consumers will be annoyed is silly. If regular consumers like AI images, they can just type "sexy anime girl" into the AI of their choice. Individual AI generated images, even 10,000 generated images literally have no value or worth, even if you like AI and are excited for the entirety of art and animation to be automated, you can agree that AI images are worthless.
|
||||
ViviP
Posts: 73 |
|
|||
Of course regular consumers wouldn’t be annoyed, it’s the user complaints that got generated images banned from Fanbox. Pixiv even apologised that it took them so long to deal with the problem. |
||||
maximilianjenus
Posts: 2866 |
|
|||
for general information, i will make a post with differenr ai workflows i qhvw witnessed in the past few months.
for transparency, i have been doing drawn art on the side for 20+ years, really minor 3dmodeling and animation, pixel art as well, with a short professional stint; but.my day job is very unrelated. each paragraph will be a different workflow, ordered by effort. type a simple prompt " fate stay night aerial , forest, nsfw" hit generate, enjoy the results. this is for newbies, takes a few minutes. type a simple prompt, it looks nothing like you want, the you start refining the prompt little by little until you get something close.enough to what you want. usually used for original or low data characters,.while getting a single image might take a few hours, people can spend weeks or doing this to develop good prompting skills. type a prompt, it looks a lot like you want,.the. send the image to an image editor t program and fix it up so it looks like you want , examples are when you get 3xtra apendages, you delete them here's or the colours are not the tone you wanted,.so you do colour adjustment, etc... a single image can take high minutes, low hours, You need drawing/image editing skills here. create a guide image starting from you own sketch or q refined prompt, as it was s ketch the amount of effort is about similar, using that guise image feed it to the ai so it create something close to what you want. then take that image to an image editor fix it up a bit , add diffusion clues, then send it back to the ai so it keeps on polishing it, the final step is to manually oldish the image. this requires a decent amount of skill both in art and in understanding. how the ai works to do the diffusion clues, the time spent is directly proportional to the complexity of what you want, it can be as low as 30 minutes , for outfit alterations to hours to assemble a complex image wit multiple charcaters. do a drawing from scratch, use diffusion clues right away for stuff you don't.care about drawing. an example is a cyberpunk anime girl, you draw the character from scratch up to a finished state, but she is in a cyberpunk layout and holding a sword. so the background is just hinted so the ai draws it ( for example , you shape the buildings and the moon) and it's the same for the sword. it.takes you as long as the original drawing takes plus 30 minutes to two hours to get the ai elements right. while i wanted to focus purely on effort, I also added skills/knowledge to the post as a last artist can spend 30 minutes In a drawing and a slower( less skilled)!artist can do the same.drawing in 8 hours . so EFFORT is effort x knowledge, because otherwise we minimize all the effort spent just getting the skills |
||||
@ASAnime6
Posts: 387 |
|
|||
why not just use it to help your art or something. not like all other users have same equal tools and money to have a fair competition.
strange to see such hate to new knowledge / methods that are meant to make humanity's focus on other stuff / make life easier. |
||||
NeverConvex
Subscriber
Posts: 2313 |
|
|||
This thread's a little old, but I thought this was an interesting development worth highlighting:
https://www.newyorker.com/culture/infinite-scroll/is-ai-art-stealing-from-artists?mbid=social_facebook&utm_brand=tny&utm_social-type=owned Relevant snippet: Last month, McKernan joined a class-action lawsuit with two other artists, Sarah Andersen and Karla Ortiz, filed by the attorneys Matthew Butterick and Joseph Saveri, against Midjourney and two other A.I. imagery generators, Stable Diffusion and DreamUp. (Other tools, such as dall-e, run on the same principles.) All three models make use of laion-5B, a nonprofit, publicly available database that indexes more than five billion images from across the Internet, including the work of many artists. The alleged wrongdoing comes down to what Butterick summarized to me as “the three ‘C’s”: The artists had not consented to have their copyrighted artwork included in the laion database; they were not compensated for their involvement, even as companies including Midjourney charged for the use of their tools; and their influence was not credited when A.I. images were produced using their work. When producing an image, these generators “present something to you as if it’s copyright free,” Butterick told me, adding that every image a generative tool produces “is an infringing, derivative work.” I've been wondering for years how long it would take for lawyers/aggrieved-concerned parties to bring (especially class-action) lawsuits that would force the courts to determine whether large-scale scraping of artificial neural network inputs is fair use, a copyright violation, etc. Back when the original neural network style transfer paper came out, I had considered playing around with it in some indie game dev stuff, but got distracted by trying to figure out whether there was any relevant case law. And, ultimately, was really surprised to find there seemed to have been almost no cases even remotely touching on the issue. I guess success enough to competitively threaten traditional artists was the push needed for cases to start bubbling up. |
||||
All times are GMT - 5 Hours |
||
|
Powered by phpBB © 2001, 2005 phpBB Group